KeyNeg in Action

import pandas as pd

# Load the Glassdoor reviews dataset
df = pd.read_csv('glassdoor_reviews.csv', encoding='latin-1')

# Define tech companies to analyze
tech_companies = ['ibm', 'oracle', 'microsoft', 'google', 'apple']

# Filter for tech companies (case-insensitive)
tech_df = df[df['firm'].str.lower().isin(tech_companies)]

# Get negative reviews (1-2 star ratings)
negative_reviews = tech_df[tech_df['overall_rating'] <= 2]

# Extract the "cons" column - this is what we'll analyze
cons_text = negative_reviews[['firm', 'cons']].dropna()

print(f"Found {len(cons_text)} negative reviews to analyze")

# For open source version
from keyneg import KeyNeg

# Or for enterprise version (air-gapped, no internet required)
# from keyneg_enterprise import KeyNeg

# Initialize the analyzer
kn = KeyNeg()

# KeyNeg comes with 95+ built-in sentiment labels including:
# - incompetent management
# - no growth opportunities
# - hostile work environment
# - layoffs
# - work life imbalance
# - and many more...

print("KeyNeg initialized successfully")

# Sample review from IBM employee
review = """Management is completely out of touch with reality.
Constant layoffs create fear and uncertainty. No clear career path
and promotions are based on politics, not merit."""

# Analyze the review
result = kn.analyze(review)

# View the results
print("Top Sentiment:", result['top_sentiment'])
print("Negativity Score:", result['negativity_score'])
print("All Sentiments:", result['sentiments'])

# Sample 200 reviews for analysis
sample_reviews = cons_text.sample(n=200, random_state=42)

# Convert to list for batch processing
review_texts = sample_reviews['cons'].tolist()

# Analyze all reviews in batch
results = kn.analyze_batch(review_texts)

# Count sentiment occurrences
from collections import Counter

all_sentiments = []
for r in results:
    all_sentiments.extend(r['sentiments'])

sentiment_counts = Counter(all_sentiments)

# Display top 10 sentiments
print("Top 10 Negative Sentiments:")
for sentiment, count in sentiment_counts.most_common(10):
    print(f"  {sentiment}: {count}")

# Add company info back to results
sample_reviews = sample_reviews.reset_index(drop=True)

# Analyze by company
company_results = {}

for company in tech_companies:
    # Get reviews for this company
    mask = sample_reviews['firm'].str.lower() == company
    company_reviews = sample_reviews[mask]['cons'].tolist()

    if len(company_reviews) > 0:
        # Analyze reviews
        company_analysis = kn.analyze_batch(company_reviews)

        # Calculate average negativity
        avg_neg = sum(r['negativity_score'] for r in company_analysis) / len(company_analysis)

        # Count sentiments
        sentiments = []
        for r in company_analysis:
            sentiments.extend(r['sentiments'])

        company_results[company] = {
            'count': len(company_reviews),
            'avg_negativity': round(avg_neg, 2),
            'top_sentiments': Counter(sentiments).most_common(3)
        }

# Display results
for company, data in company_results.items():
    print(f"\n{company.upper()}")
    print(f"  Reviews: {data['count']}")
    print(f"  Avg Negativity: {data['avg_negativity']}")
    print(f"  Top Issues: {data['top_sentiments']}")